verification-before-completion
Verify Before Claiming Completion
Also available from: Dimon94,Cycleaddict,Cygnusfear,DYAI2025,davila7,CodingCossack,ANcpLua,obra
This skill prevents AI agents from making false completion claims by requiring actual verification evidence before reporting success. It ensures tests pass, builds succeed, and work is genuinely complete.
Download the skill ZIP
Upload in Claude
Go to Settings → Capabilities → Skills → Upload skill
Toggle on and start using
Test it
Using "verification-before-completion". The tests should pass now
Expected outcome:
Running test suite...
Results: 42 tests, 42 passed, 0 failed
Exit code: 0
All tests pass.
Using "verification-before-completion". The build completed successfully
Expected outcome:
Running build command...
> npm run build
Exit code: 0
Build completed successfully.
Output length: 2.3MB
Using "verification-before-completion". The bug is fixed
Expected outcome:
Running reproduction test...
Original failing input: {id: 123}
Expected: Error thrown
Actual: Error thrown
Bug is confirmed fixed.
Security Audit
Low RiskThis is a meta-skill providing verification guidelines. The static scanner flagged 13 external_commands patterns and 1 weak crypto pattern, but evaluation shows these are FALSE POSITIVES: the skill contains instructions for AI agents to run verification commands (tests, linters, builds) as text examples - not actual executable code. No cryptographic code exists. System/network reconnaissance flags refer to standard DevOps terms (VCS, git diff). This skill promotes honest verification practices and poses no security risk.
High Risk Issues (1)
Medium Risk Issues (1)
Low Risk Issues (1)
Quality Score
What You Can Build
Test Verification Before Commit
Before committing code, require the AI to run test commands and show actual pass/fail counts from fresh execution.
Build Verification Before PR
Prevent premature PR creation by requiring build command output showing exit code 0 before claiming success.
Regression Test Validation
Ensure regression tests follow red-green pattern: pass after fix, fail without fix, pass after restore.
Try These Prompts
Before claiming this work is complete, run the verification command and show me the actual output. Do not claim success without evidence.
Run the full test suite and report: total tests, passed, failed, and the exact error messages for any failures. Show the exit code.
Run the build command and show the exit code. Report any compilation errors or warnings from the output.
Review the original requirements. For each requirement, state: met/unmet/gap. Cite specific evidence from the code or test output.
Best Practices
- Always run fresh verification commands - never trust cached or previous results
- Read the full output including exit codes before making any claims
- Express actual status with evidence, not confidence or assumptions
Avoid
- Claiming completion based on agent success reports without independent verification
- Using words like 'should', 'probably', 'seems to' before verification
- Relying on partial checks like linter passing when full build verification is needed